[FEATURE] Sync files between Claude Code and Claude web project knowledge

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Claude Code and Claude web projects serve complementary roles (Code for execution/data, web for strategy/writing), but there's no programmatic way to sync files between them. When Claude Code generates
analysis outputs, session briefings, or corrects stale data in project files, the only way to get them into Claude web's project knowledge is manual drag-and-drop through the browser. The human becomes
the bottleneck in an otherwise automated workflow, and files silently drift out of sync when the manual step gets forgotten.

Proposed Solution

A CLI command that can push/pull files to/from a Claude web project's knowledge store:

claude project upload --project <id> path/to/file.md
claude project list --project <id>
claude project sync --project <id> path/to/directory/

Authentication could reuse the existing Anthropic API key or OAuth flow. Even read-only access (listing what's currently in project knowledge) would help Claude Code know what's stale.

Alternative Solutions

• A shared filesystem or watched directory that both Claude Code and Claude web can access
• A webhook or event that triggers on Claude Code session end, pushing designated files to a project
• MCP server integration that exposes project knowledge as a tool
• Manual workaround (current): generate a single CURRENT-STATE.md briefing at end of each session to minimize uploads to one file — still requires manual browser upload

Priority

High - Significant impact on productivity

Feature Category

API and model interactions

Use Case Example

I run a research project where Claude Code processes databases, runs statistical analyses, and generates reports. Claude web handles strategic planning and article drafting. At the end of each Claude
Code session, a script generates a briefing file with current numbers, findings status, and session notes. Today I have to manually download that file, open the Claude web project, and upload it before
my next conversation there. If I forget (or if Claude Code corrected a number mid-session), Claude web works with stale data and produces documents with wrong figures — which I then have to catch and
fix manually. A claude project upload command at end-of-session would close this loop entirely.

Additional Context

The building blocks are close — Claude Code already has claude CLI commands, API authentication, and file generation. Claude web already has project knowledge file management. The gap is just the bridge
between them. Even a minimal v1 (upload a single file by project ID) would eliminate the most painful part of the workflow.